Introduction

Aether is a cloud development platform where an AI agent writes code and tests applications in a real browser — all from natural language. Instead of switching between your IDE, terminal, cloud console, and browser, you describe what you want and Aether’s agent handles the rest inside a cloud workspace.

​

Key Features

​

How It Works

1. Create a project — Start from scratch or link a GitHub repository
2. Describe your task — Tell the agent what to build, fix, or change
3. Watch it work — The agent writes code, runs commands, and tests in a real browser
4. Review and ship — Inspect changes, preview the result, and open a pull request

​

How It’s Structured

A few facts shape how Aether works day to day:

• One task = one agent = one workspace. Every task is bound to exactly one agent run inside exactly one isolated workspace VM. Work never bleeds across tasks. See tasks and lifecycle.
• You pick the agent. Aether ships three selectable agent providers — Codex (default), Claude Code, and OpenCode. Capabilities and behavior vary by provider. See agents and models.
• Connect a provider to run tasks. Codex is the default agent, but the chosen provider runs under your own connected account — connect one in Settings → Connections before your first task. See connect agent providers.
• You’re billed for VM runtime, not tokens. Usage is metered in credits, where credits track how long your workspace VMs run. LLM tokens are captured for usage reporting but never billed. See billing and credits.

​

Why It Matters

​

The agent runs your code, not just writes it

Most AI coding tools generate code and hope it works. Aether’s agent actually executes it — starts dev servers, runs tests, opens a browser, and verifies the result before saying it’s done. When it writes a fix, it runs the test suite. When it builds a feature, it opens it in a browser and checks that it renders correctly. You get working code, not code that looks right.

​

Every task gets a full environment

Each workspace is a full Linux machine with persistent storage. The agent has a real terminal, real file system, and a real browser. It can install dependencies, start databases, run build tools — anything you’d do locally. This isn’t a stateless sandbox that resets between prompts. Your workspace persists across your task’s turns — it suspends when idle and resumes fast, so you can come back later and continue exactly where the task left off. New tasks start immediately when a warm pool has a ready machine for the project and requested size; otherwise Aether cold-starts the requested size.

​

Code isolation with actual execution

Git worktrees and branch-based tools give you code isolation, but the agent still can’t run anything. It writes code blind and hands it back to you to verify. Aether gives you both isolation and execution. The agent catches runtime errors, broken imports, type failures, and failing tests before you ever look at the code. The difference is getting a PR that works vs a PR that compiles.

​

Next Steps